Vast.ai

Flotera uses the price per GPU-hour to compare and set Vast.ai prices. Some requirements depend on the number of GPUs in the server: memory, disk, and ports.

Connecting

1. Get an API key in your Vast.ai account.
2. Add it in Marketplaces in Flotera. Existing listings and rentals appear after synchronization, usually within 1–2 minutes.
3. Install the Vast.ai agent yourself, following the Vast.ai host setup. In Platforms on the server card, run Autodetect and Link to a machine to select your Vast.ai machine through its API.
4. Run the readiness check, fix the errors, and run the check again.
5. Set a price and enable showing in search.

Verification

Vast.ai checks the server separately and assigns it a verification status. The Flotera check helps prepare the server but does not replace the marketplace’s decision. The conditions are described in the Vast.ai documentation.

Before preparing hardware, check CPU compatibility, VRAM size, the supported CUDA version, and the other requirements on the Platforms tab. Hardware limitations cannot be fixed by configuring the agent.

Pricing

The panel shows your price and the market metrics for the same GPU model. When comparing, consider the number of offers and the composition of the sample.

Market prices are available after you connect your Vast.ai account. See market prices for help comparing offers.

Confirm price changes in the panel. If a change is still pending, wait for the result before trying again.

Entering and leaving search

Publish creates the Vast listing. Withdraw removes the listing; Vast does not keep a hidden listing with the old terms. During an automatic pause, Flotera withdraws the Vast.ai listing and restores it with its previous terms when the pause ends. Choose Don’t bring it back to leave it unpublished.

Before maintenance, withdraw the listing and check that there is no active rental.

Publishing does not guarantee a rental

If the server is visible in search but is not rented, check the price, specifications, reliability, and demand for this GPU model. Comparison with the median is one reference point.

A published listing may not be available to renters yet. Flotera checks the public Vast.ai catalog and shows such an offer as Indexing; after an outage this takes up to an hour. If renters still do not see it after an hour, the offer becomes Hidden and you get a notification. When Vast.ai has not heard the server for 10 minutes, the marketplace status becomes Offline — see server states.

The arrow next to reliability shows whether the rating has risen or fallen; its tooltip shows the period. If reliability is falling, check the server’s connection and recent outages.

CUDA compatibility

The driver version determines which CUDA versions the renter’s applications can use. A newer compatible driver expands the choice of images but does not guarantee demand for the server. The CUDA field in nvidia-smi shows the version supported by the driver, not whether CUDA Toolkit is installed on the host. For details, see the nvidia-smi reference.

Update the driver between rentals: the GPUs become temporarily unavailable, and a reboot may be required.

Market pricesCompare your price with the market.